Weill Cornell models which protein degraders will work before the chemistry starts

Degradation and inhibition price binding affinity differently, and that is what the turnover result is really about. An inhibitor works while it is sitting on its target, so its effect tracks occupancy ^[12]. A degrader is a two-ended molecule that holds the target against the cell's own disposal machinery ^[6], redirecting equipment the cell already runs ^[20]; once the target has been destroyed, letting go costs little, because the cell must synthesise a replacement before the target exists again. The steady-state level therefore depends on rebuild rate as much as on grip, and Du's finding is the quantitative form of that: slow-turnover targets can be potently degraded by relatively weak binders ^[10]. Measure turnover first and you know how much affinity you actually need to pay a chemistry team to find.

Elemento's illustration of why you would want destruction rather than silencing is a protein sitting at 200 copies in a healthy cell and 2,000 after a cancer-linked mutation ^[13], a tenfold excess ^[14] that switching the protein off leaves in place. Alongside the framework, the group reports a set of proteins it judges high-value for degrader development ^[11], and a route through optimisation it describes as the most cost-effective one ^[9].

The thing the announcement does not tell you is how the model performed on degraders it had not already seen. The phys.org account gives no accuracy figures and no count of the targets on that list, and it does not describe a prospective experimental test ^[16]. The claim that the framework could shave years off a development program is the institution's projection rather than a measured outcome ^[15]. That evidence, if it exists, sits in the paper itself ^[3].

Moving an empirical screen into simulation saves money only when the simulation's inputs cost less than the screen. Here the inputs are described as easily obtained laboratory measurements ^[8], and that description is load-bearing for the entire argument: if pinning down a target's turnover rate in the relevant cell type becomes its own multi-month project, the expense relocates rather than vanishes. Set against that, what is being displaced is repeated trial and error, which is where degrader programs currently burn time and reagents ^[7].

Du frames the nearer-term use as triage between modalities, choosing among gene therapy, mRNA, a degrader and an antibody-drug conjugate for a given patient ^[18], and the lab has already published a companion model aimed at predicting drug safety ahead of clinical trials ^[17]. Elemento's stated goal is a drug customised to one unique patient rather than one designed for a thousand ^[19]. A degradability ranking is a defensible first instrument for that, on the condition that the ranking holds for degraders nobody has synthesised yet.